Simulating Head Motion in a Planning CT for Data Augmentation of Head and Neck Based Deep-Learning Networks
M Gardner1*, Y Ben Bouchta1, J Sykes2, P Keall1, (1) University of Sydney, Sydney, NSW, AU (2) Blacktown Hospital, Blacktown, NSW ,AU,
Presentations
PO-GePV-M-160 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: The purpose of this work was to create a method to create realistic synthetic deformations, which can be applied to a planning CT to create multiple volumes to be used to increase the quality and quantity of data used for deep-learning training. Deep-learning methods for detecting head and neck (H&N) tumor motion only have data from one planning CT volume to train on, unlike for respiratory motion in which multiple volumes are acquired in a planning 4DCT. To accurately train H&N deep-learning networks to be robust to motion a data augmentation method is needed to replicate H&N motion.
Methods: The literature was investigated to determine the types of H&N motion that would be typical for a H&N patient during radiotherapy. Work on the biomechanical kinematics of the H&N were investigated, as well as results showing the internal H&N motion during radiotherapy.
Results: From this analysis common H&N motion was categorized as either head rotation or internal tissue motion. The head rotation was achieved by rotating the head volume, and was split into anterior-posterior, superior-inferior and left-right rotation. The center points of these rotations were the gaps between cervical vertebrae. The internal tissue motion was achieved by shifting the contoured GTV, with a gaussian smoothing filter applied to the deformation. Deformations were created by adding the head rotation motion to the internal tissue motion. A rigid/non-rigid registration between the planning CT and the deformed CT was used as a final processing step.
Conclusion: Synthetic deformations were created that can be applied to H&N planning CTs. These deformations and magnitudes were based on literature detailing H&N motion.
Funding Support, Disclosures, and Conflict of Interest: This project is funded by Cancer Australia, funded by the Australian Government
Keywords
CT, Deformation, Patient Movement
Taxonomy
Not Applicable / None Entered.
Contact Email